Optimize ulong/double Vector conversion on xarch with AVX512VL instructions.

[kote01-coder]

With some of the new AVX512VL instructions, we can optimize Vector uint32/uint64 to/from Vector single/double conversions within the JIT.

These would be Vector128/Vector256 ConvertToInt64, ConvertToInt32 etc. optimizations.

vcvtpd2qq – faster packed double to int64 conversions (also vcvttpd2qq)
vcvtpd2udq – faster packed double to uint32 conversions (also vcvttpd2udq)
vcvtpd2uqq – faster packed double to uint64 conversions (also vcvttpd2uqq)
vcvtps2qq – faster packed single to int64 conversions (also vcvttps2qq)
vcvtps2udq – faster packed single to uint32 conversions (also vcvttps2udq)
vcvtps2uqq – faster packed single to uint64 conversions (also vcvttps2uqq)

These would be Vector128/Vector256 ConvertToDouble and ConvertToSingle optimizations.

vcvtqq2pd – faster packed int64 to double conversions
vcvtqq2ps – faster packed int64 to single conversions
vcvtudq2pd – faster packed uint32 to double conversions
vcvtudq2ps – faster packed uint32 to single conversions
vcvtuqq2pd – faster packed uint64 to double conversions
vcvtuqq2ps – faster packed uint64 to single conversions

As it stands, I believe many of these conversions are done with with a software fallback or a sequence of more complicated instructions, whereas with AVX512VL we can use a single instruction to perform the conversions.